Support non-key filtered indexes, fluent fallback modes and metadata round-trips

[jogibear9988]

Fixes #199.

A unique index on IpaUserIdentifier filtered by IpaUserIdentifier IS NOT NULL AND Archive = 0 previously failed because shared validation required every filter column to be an index key, and null filter values were unsupported. SQL Server, PostgreSQL and SQLite now support this definition through both the classic and fluent APIs, and can read back the supported index definition for recreation.

Changes

• Validate filter columns against table columns and render null/DBNull.Value comparisons as IS NULL or IS NOT NULL.
• Add Index.UnsupportedFilterBehavior with UnsupportedIndexFilterBehavior.Throw (default) and Ignore, exposed through fluent .OnUnsupportedFilter(...).
• Apply the policy to providers without implemented filter support, the SQL Server 2005 dialect, and unsupported cases of Oracle's existing expression emulation. Supported providers still apply filters in Ignore mode; other invalid options and database errors are not suppressed.
• Share filter formatting and catalog parsing. Read back null checks, non-key predicates, quoted identifiers and escaped strings, including strings containing AND.
• Preserve keys, included columns and flags through metadata and fluent snapshots. Scope PostgreSQL metadata to the requested table, preserve column order, and emit INCLUDE before WHERE.
• Match SQLite index SQL by exact name and preserve filters through table reconstruction. Test recreating an index from its read-back definition through classic and fluent APIs.
• Update the manual, generated documentation and runner guide.

Usage

migration.Create.Index("UX_ActiveUsers").OnTable("Users")
    .WithColumns("IpaUserIdentifier").Unique()
    .WithFilter(
        new FilterItem { ColumnName = "IpaUserIdentifier", Filter = FilterType.NotEqualTo, Value = null },
        new FilterItem { ColumnName = "Archive", Filter = FilterType.EqualTo, Value = 0 })
    .OnUnsupportedFilter(UnsupportedIndexFilterBehavior.Throw);

Read back with Database.GetIndexes("Users") or Schema.Table("Users").Indexes() in fluent migrations.

Behavior and limits

Ignore creates an unfiltered index. For a unique index, uniqueness then applies to all rows. The fallback policy is an authoring option and is not stored in database metadata.

Oracle retains its limited non-unique, key-column expression emulation; expression read-back as FilterItems remains unsupported. SQL preview still rejects filtered indexes, including Ignore mode. Predicates that cannot be represented by FilterItems, such as OR, fail explicitly.

Validation

Rebased onto current master and verified:

• dotnet build Migrator.slnx --no-restore -v quiet: passed.
• Unit suite: 799 passed, 0 skipped.
• SQLite suite: 421 passed, 0 skipped.
• Documentation generation consistency, local links, anchors and paired sample structure: passed.
• Diff whitespace check with the repository's CRLF content accounted for: passed.
• Live SQL Server/PostgreSQL round-trip tests added for CI; not run locally because the servers are unavailable. Local SQL generation and catalog-reader regression tests passed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.